Analytics questions turn into several calls — totals, then paths, then referrers for the interesting ones — which is exactly the pattern that trips an API limit and leaves the assistant with a half-answer. Retrying on 429 while respecting the reset header means the chain finishes instead of failing midway. Goatcounter's own privacy stance is the other reason to use it: the numbers come from a service that is not tracking your visitors individually.
A client for Goatcounter, the privacy-focused web analytics service, covering most of its API. It reads totals and hit series, and breaks traffic down by referrer, browser, operating system, screen size and location. The client initialises lazily, so tools can be listed before credentials are configured.
- `Goatcounter_get_stats_total` and `Goatcounter_get_stats_hits` return overall totals and the hit series
- `Goatcounter_get_stats_refs` breaks traffic down by referrer
- `Goatcounter_get_stats_browsers` and `Goatcounter_get_stats_systems` break it down by browser and operating system
- `Goatcounter_get_stats_sizes` and `Goatcounter_get_stats_locations` cover screen size and geography
- `Goatcounter_list_paths` lists the pages being counted, and `Goatcounter_list_sites` the sites on the account
- `Goatcounter_get_me` reports the authenticated account
- Automatic retries on HTTP 429, honouring `X-Rate-Limit-Reset` when present and falling back to exponential backoff with jitter, up to 5 attempts
A Goatcounter site and an API token generated under Settings then API tokens, with the permissions matching what you plan to call. Set `GOATCOUNTER_CODE` to your site code — the subdomain part — and `GOATCOUNTER_API_KEY` to the token. Install with `pip install goatcounter-mcp-server` (package `goatcounter-mcp-server`, version 0.2.4) and run the `goatcounter-mcp-server` command, or install it for Claude Desktop through the Smithery CLI.
One command plus a key — pip install goatcounter-mcp-server, then supply credentials